2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * The IP fragmentation functionality.
8 * Authors: Fred N. van Kempen <waltje@uWalt.NL.Mugnet.ORG>
9 * Alan Cox <alan@lxorguk.ukuu.org.uk>
12 * Alan Cox : Split from ip.c , see ip_input.c for history.
13 * David S. Miller : Begin massive cleanup...
14 * Andi Kleen : Add sysctls.
15 * xxxx : Overlapfrag bug.
16 * Ultima : ip_expire() kernel panic.
17 * Bill Hawes : Frag accounting and evictor fixes.
18 * John McDonald : 0 length frag bug.
19 * Alexey Kuznetsov: SMP races, threading, cleanup.
20 * Patrick McHardy : LRU queue of frag heads for evictor.
23 #define pr_fmt(fmt) "IPv4: " fmt
25 #include <linux/compiler.h>
26 #include <linux/module.h>
27 #include <linux/types.h>
29 #include <linux/jiffies.h>
30 #include <linux/skbuff.h>
31 #include <linux/list.h>
33 #include <linux/icmp.h>
34 #include <linux/netdevice.h>
35 #include <linux/jhash.h>
36 #include <linux/random.h>
37 #include <linux/slab.h>
38 #include <net/route.h>
43 #include <net/checksum.h>
44 #include <net/inetpeer.h>
45 #include <net/inet_frag.h>
46 #include <linux/tcp.h>
47 #include <linux/udp.h>
48 #include <linux/inet.h>
49 #include <linux/netfilter_ipv4.h>
50 #include <net/inet_ecn.h>
52 /* NOTE. Logic of IP defragmentation is parallel to corresponding IPv6
53 * code now. If you change something here, _PLEASE_ update ipv6/reassembly.c
54 * as well. Or notify me, at least. --ANK
57 static int sysctl_ipfrag_max_dist __read_mostly = 64;
61 struct inet_skb_parm h;
65 #define FRAG_CB(skb) ((struct ipfrag_skb_cb *)((skb)->cb))
67 /* Describe an entry in the "incomplete datagrams" queue. */
69 struct inet_frag_queue q;
76 u8 ecn; /* RFC3168 support */
79 struct inet_peer *peer;
82 static inline u8 ip4_frag_ecn(u8 tos)
84 return 1 << (tos & INET_ECN_MASK);
87 static struct inet_frags ip4_frags;
89 int ip_frag_nqueues(struct net *net)
91 return net->ipv4.frags.nqueues;
94 int ip_frag_mem(struct net *net)
96 return sum_frag_mem_limit(&net->ipv4.frags);
99 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
100 struct net_device *dev);
102 struct ip4_create_arg {
107 static unsigned int ipqhashfn(__be16 id, __be32 saddr, __be32 daddr, u8 prot)
109 net_get_random_once(&ip4_frags.rnd, sizeof(ip4_frags.rnd));
110 return jhash_3words((__force u32)id << 16 | prot,
111 (__force u32)saddr, (__force u32)daddr,
112 ip4_frags.rnd) & (INETFRAGS_HASHSZ - 1);
115 static unsigned int ip4_hashfn(struct inet_frag_queue *q)
119 ipq = container_of(q, struct ipq, q);
120 return ipqhashfn(ipq->id, ipq->saddr, ipq->daddr, ipq->protocol);
123 static bool ip4_frag_match(struct inet_frag_queue *q, void *a)
126 struct ip4_create_arg *arg = a;
128 qp = container_of(q, struct ipq, q);
129 return qp->id == arg->iph->id &&
130 qp->saddr == arg->iph->saddr &&
131 qp->daddr == arg->iph->daddr &&
132 qp->protocol == arg->iph->protocol &&
133 qp->user == arg->user;
136 static void ip4_frag_init(struct inet_frag_queue *q, void *a)
138 struct ipq *qp = container_of(q, struct ipq, q);
139 struct netns_ipv4 *ipv4 = container_of(q->net, struct netns_ipv4,
141 struct net *net = container_of(ipv4, struct net, ipv4);
143 struct ip4_create_arg *arg = a;
145 qp->protocol = arg->iph->protocol;
146 qp->id = arg->iph->id;
147 qp->ecn = ip4_frag_ecn(arg->iph->tos);
148 qp->saddr = arg->iph->saddr;
149 qp->daddr = arg->iph->daddr;
150 qp->user = arg->user;
151 qp->peer = sysctl_ipfrag_max_dist ?
152 inet_getpeer_v4(net->ipv4.peers, arg->iph->saddr, 1) : NULL;
155 static __inline__ void ip4_frag_free(struct inet_frag_queue *q)
159 qp = container_of(q, struct ipq, q);
161 inet_putpeer(qp->peer);
165 /* Destruction primitives. */
167 static __inline__ void ipq_put(struct ipq *ipq)
169 inet_frag_put(&ipq->q, &ip4_frags);
172 /* Kill ipq entry. It is not destroyed immediately,
173 * because caller (and someone more) holds reference count.
175 static void ipq_kill(struct ipq *ipq)
177 inet_frag_kill(&ipq->q, &ip4_frags);
180 /* Memory limiting on fragments. Evictor trashes the oldest
181 * fragment queue until we are back under the threshold.
183 static void ip_evictor(struct net *net)
187 evicted = inet_frag_evictor(&net->ipv4.frags, &ip4_frags, false);
189 IP_ADD_STATS_BH(net, IPSTATS_MIB_REASMFAILS, evicted);
193 * Oops, a fragment queue timed out. Kill it and send an ICMP reply.
195 static void ip_expire(unsigned long arg)
200 qp = container_of((struct inet_frag_queue *) arg, struct ipq, q);
201 net = container_of(qp->q.net, struct net, ipv4.frags);
203 spin_lock(&qp->q.lock);
205 if (qp->q.last_in & INET_FRAG_COMPLETE)
210 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMTIMEOUT);
211 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
213 if ((qp->q.last_in & INET_FRAG_FIRST_IN) && qp->q.fragments != NULL) {
214 struct sk_buff *head = qp->q.fragments;
215 const struct iphdr *iph;
219 head->dev = dev_get_by_index_rcu(net, qp->iif);
223 /* skb has no dst, perform route lookup again */
225 err = ip_route_input_noref(head, iph->daddr, iph->saddr,
226 iph->tos, head->dev);
231 * Only an end host needs to send an ICMP
232 * "Fragment Reassembly Timeout" message, per RFC792.
234 if (qp->user == IP_DEFRAG_AF_PACKET ||
235 (qp->user == IP_DEFRAG_CONNTRACK_IN &&
236 skb_rtable(head)->rt_type != RTN_LOCAL))
240 /* Send an ICMP "Fragment Reassembly Timeout" message. */
241 icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
246 spin_unlock(&qp->q.lock);
250 /* Find the correct entry in the "incomplete datagrams" queue for
251 * this IP datagram, and create new one, if nothing is found.
253 static inline struct ipq *ip_find(struct net *net, struct iphdr *iph, u32 user)
255 struct inet_frag_queue *q;
256 struct ip4_create_arg arg;
262 read_lock(&ip4_frags.lock);
263 hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol);
265 q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash);
266 if (IS_ERR_OR_NULL(q)) {
267 inet_frag_maybe_warn_overflow(q, pr_fmt());
270 return container_of(q, struct ipq, q);
273 /* Is the fragment too far ahead to be part of ipq? */
274 static inline int ip_frag_too_far(struct ipq *qp)
276 struct inet_peer *peer = qp->peer;
277 unsigned int max = sysctl_ipfrag_max_dist;
278 unsigned int start, end;
286 end = atomic_inc_return(&peer->rid);
289 rc = qp->q.fragments && (end - start) > max;
294 net = container_of(qp->q.net, struct net, ipv4.frags);
295 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
301 static int ip_frag_reinit(struct ipq *qp)
304 unsigned int sum_truesize = 0;
306 if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) {
307 atomic_inc(&qp->q.refcnt);
311 fp = qp->q.fragments;
313 struct sk_buff *xp = fp->next;
315 sum_truesize += fp->truesize;
319 sub_frag_mem_limit(&qp->q, sum_truesize);
324 qp->q.fragments = NULL;
325 qp->q.fragments_tail = NULL;
332 /* Add new segment to existing queue. */
333 static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
335 struct sk_buff *prev, *next;
336 struct net_device *dev;
342 if (qp->q.last_in & INET_FRAG_COMPLETE)
345 if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
346 unlikely(ip_frag_too_far(qp)) &&
347 unlikely(err = ip_frag_reinit(qp))) {
352 ecn = ip4_frag_ecn(ip_hdr(skb)->tos);
353 offset = ntohs(ip_hdr(skb)->frag_off);
354 flags = offset & ~IP_OFFSET;
356 offset <<= 3; /* offset is in 8-byte chunks */
357 ihl = ip_hdrlen(skb);
359 /* Determine the position of this fragment. */
360 end = offset + skb->len - ihl;
363 /* Is this the final fragment? */
364 if ((flags & IP_MF) == 0) {
365 /* If we already have some bits beyond end
366 * or have different end, the segment is corrupted.
368 if (end < qp->q.len ||
369 ((qp->q.last_in & INET_FRAG_LAST_IN) && end != qp->q.len))
371 qp->q.last_in |= INET_FRAG_LAST_IN;
376 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
377 skb->ip_summed = CHECKSUM_NONE;
379 if (end > qp->q.len) {
380 /* Some bits beyond end -> corruption. */
381 if (qp->q.last_in & INET_FRAG_LAST_IN)
390 if (pskb_pull(skb, ihl) == NULL)
393 err = pskb_trim_rcsum(skb, end - offset);
397 /* Find out which fragments are in front and at the back of us
398 * in the chain of fragments so far. We must know where to put
399 * this fragment, right?
401 prev = qp->q.fragments_tail;
402 if (!prev || FRAG_CB(prev)->offset < offset) {
407 for (next = qp->q.fragments; next != NULL; next = next->next) {
408 if (FRAG_CB(next)->offset >= offset)
414 /* We found where to put this one. Check for overlap with
415 * preceding fragment, and, if needed, align things so that
416 * any overlaps are eliminated.
419 int i = (FRAG_CB(prev)->offset + prev->len) - offset;
427 if (!pskb_pull(skb, i))
429 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
430 skb->ip_summed = CHECKSUM_NONE;
436 while (next && FRAG_CB(next)->offset < end) {
437 int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */
440 /* Eat head of the next overlapped fragment
441 * and leave the loop. The next ones cannot overlap.
443 if (!pskb_pull(next, i))
445 FRAG_CB(next)->offset += i;
447 if (next->ip_summed != CHECKSUM_UNNECESSARY)
448 next->ip_summed = CHECKSUM_NONE;
451 struct sk_buff *free_it = next;
453 /* Old fragment is completely overridden with
461 qp->q.fragments = next;
463 qp->q.meat -= free_it->len;
464 sub_frag_mem_limit(&qp->q, free_it->truesize);
469 FRAG_CB(skb)->offset = offset;
471 /* Insert this fragment in the chain of fragments. */
474 qp->q.fragments_tail = skb;
478 qp->q.fragments = skb;
482 qp->iif = dev->ifindex;
485 qp->q.stamp = skb->tstamp;
486 qp->q.meat += skb->len;
488 add_frag_mem_limit(&qp->q, skb->truesize);
490 qp->q.last_in |= INET_FRAG_FIRST_IN;
492 if (ip_hdr(skb)->frag_off & htons(IP_DF) &&
493 skb->len + ihl > qp->q.max_size)
494 qp->q.max_size = skb->len + ihl;
496 if (qp->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
497 qp->q.meat == qp->q.len) {
498 unsigned long orefdst = skb->_skb_refdst;
500 skb->_skb_refdst = 0UL;
501 err = ip_frag_reasm(qp, prev, dev);
502 skb->_skb_refdst = orefdst;
507 inet_frag_lru_move(&qp->q);
516 /* Build a new IP datagram from all its fragments. */
518 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
519 struct net_device *dev)
521 struct net *net = container_of(qp->q.net, struct net, ipv4.frags);
523 struct sk_buff *fp, *head = qp->q.fragments;
532 ecn = ip_frag_ecn_table[qp->ecn];
533 if (unlikely(ecn == 0xff)) {
537 /* Make the one we just received the head. */
540 fp = skb_clone(head, GFP_ATOMIC);
544 fp->next = head->next;
546 qp->q.fragments_tail = fp;
549 skb_morph(head, qp->q.fragments);
550 head->next = qp->q.fragments->next;
552 consume_skb(qp->q.fragments);
553 qp->q.fragments = head;
556 WARN_ON(head == NULL);
557 WARN_ON(FRAG_CB(head)->offset != 0);
559 /* Allocate a new buffer for the datagram. */
560 ihlen = ip_hdrlen(head);
561 len = ihlen + qp->q.len;
567 /* Head of list must not be cloned. */
568 if (skb_unclone(head, GFP_ATOMIC))
571 /* If the first fragment is fragmented itself, we split
572 * it to two chunks: the first with data and paged part
573 * and the second, holding only fragments. */
574 if (skb_has_frag_list(head)) {
575 struct sk_buff *clone;
578 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
580 clone->next = head->next;
582 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
583 skb_frag_list_init(head);
584 for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
585 plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
586 clone->len = clone->data_len = head->data_len - plen;
587 head->data_len -= clone->len;
588 head->len -= clone->len;
590 clone->ip_summed = head->ip_summed;
591 add_frag_mem_limit(&qp->q, clone->truesize);
594 skb_push(head, head->data - skb_network_header(head));
596 sum_truesize = head->truesize;
597 for (fp = head->next; fp;) {
600 struct sk_buff *next = fp->next;
602 sum_truesize += fp->truesize;
603 if (head->ip_summed != fp->ip_summed)
604 head->ip_summed = CHECKSUM_NONE;
605 else if (head->ip_summed == CHECKSUM_COMPLETE)
606 head->csum = csum_add(head->csum, fp->csum);
608 if (skb_try_coalesce(head, fp, &headstolen, &delta)) {
609 kfree_skb_partial(fp, headstolen);
611 if (!skb_shinfo(head)->frag_list)
612 skb_shinfo(head)->frag_list = fp;
613 head->data_len += fp->len;
614 head->len += fp->len;
615 head->truesize += fp->truesize;
619 sub_frag_mem_limit(&qp->q, sum_truesize);
623 head->tstamp = qp->q.stamp;
624 IPCB(head)->frag_max_size = qp->q.max_size;
627 /* max_size != 0 implies at least one fragment had IP_DF set */
628 iph->frag_off = qp->q.max_size ? htons(IP_DF) : 0;
629 iph->tot_len = htons(len);
631 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMOKS);
632 qp->q.fragments = NULL;
633 qp->q.fragments_tail = NULL;
637 LIMIT_NETDEBUG(KERN_ERR pr_fmt("queue_glue: no memory for gluing queue %p\n"),
642 net_info_ratelimited("Oversized IP packet from %pI4\n", &qp->saddr);
644 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
648 /* Process an incoming IP datagram fragment. */
649 int ip_defrag(struct sk_buff *skb, u32 user)
654 net = skb->dev ? dev_net(skb->dev) : dev_net(skb_dst(skb)->dev);
655 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMREQDS);
657 /* Start by cleaning up the memory. */
660 /* Lookup (or create) queue header */
661 if ((qp = ip_find(net, ip_hdr(skb), user)) != NULL) {
664 spin_lock(&qp->q.lock);
666 ret = ip_frag_queue(qp, skb);
668 spin_unlock(&qp->q.lock);
673 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
677 EXPORT_SYMBOL(ip_defrag);
679 struct sk_buff *ip_check_defrag(struct sk_buff *skb, u32 user)
684 if (skb->protocol != htons(ETH_P_IP))
687 if (!skb_copy_bits(skb, 0, &iph, sizeof(iph)))
690 if (iph.ihl < 5 || iph.version != 4)
693 len = ntohs(iph.tot_len);
694 if (skb->len < len || len < (iph.ihl * 4))
697 if (ip_is_fragment(&iph)) {
698 skb = skb_share_check(skb, GFP_ATOMIC);
700 if (!pskb_may_pull(skb, iph.ihl*4))
702 if (pskb_trim_rcsum(skb, len))
704 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
705 if (ip_defrag(skb, user))
712 EXPORT_SYMBOL(ip_check_defrag);
717 static struct ctl_table ip4_frags_ns_ctl_table[] = {
719 .procname = "ipfrag_high_thresh",
720 .data = &init_net.ipv4.frags.high_thresh,
721 .maxlen = sizeof(int),
723 .proc_handler = proc_dointvec
726 .procname = "ipfrag_low_thresh",
727 .data = &init_net.ipv4.frags.low_thresh,
728 .maxlen = sizeof(int),
730 .proc_handler = proc_dointvec
733 .procname = "ipfrag_time",
734 .data = &init_net.ipv4.frags.timeout,
735 .maxlen = sizeof(int),
737 .proc_handler = proc_dointvec_jiffies,
742 static struct ctl_table ip4_frags_ctl_table[] = {
744 .procname = "ipfrag_secret_interval",
745 .data = &ip4_frags.secret_interval,
746 .maxlen = sizeof(int),
748 .proc_handler = proc_dointvec_jiffies,
751 .procname = "ipfrag_max_dist",
752 .data = &sysctl_ipfrag_max_dist,
753 .maxlen = sizeof(int),
755 .proc_handler = proc_dointvec_minmax,
761 static int __net_init ip4_frags_ns_ctl_register(struct net *net)
763 struct ctl_table *table;
764 struct ctl_table_header *hdr;
766 table = ip4_frags_ns_ctl_table;
767 if (!net_eq(net, &init_net)) {
768 table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL);
772 table[0].data = &net->ipv4.frags.high_thresh;
773 table[1].data = &net->ipv4.frags.low_thresh;
774 table[2].data = &net->ipv4.frags.timeout;
776 /* Don't export sysctls to unprivileged users */
777 if (net->user_ns != &init_user_ns)
778 table[0].procname = NULL;
781 hdr = register_net_sysctl(net, "net/ipv4", table);
785 net->ipv4.frags_hdr = hdr;
789 if (!net_eq(net, &init_net))
795 static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net)
797 struct ctl_table *table;
799 table = net->ipv4.frags_hdr->ctl_table_arg;
800 unregister_net_sysctl_table(net->ipv4.frags_hdr);
804 static void ip4_frags_ctl_register(void)
806 register_net_sysctl(&init_net, "net/ipv4", ip4_frags_ctl_table);
809 static inline int ip4_frags_ns_ctl_register(struct net *net)
814 static inline void ip4_frags_ns_ctl_unregister(struct net *net)
818 static inline void ip4_frags_ctl_register(void)
823 static int __net_init ipv4_frags_init_net(struct net *net)
825 /* Fragment cache limits.
827 * The fragment memory accounting code, (tries to) account for
828 * the real memory usage, by measuring both the size of frag
829 * queue struct (inet_frag_queue (ipv4:ipq/ipv6:frag_queue))
830 * and the SKB's truesize.
832 * A 64K fragment consumes 129736 bytes (44*2944)+200
833 * (1500 truesize == 2944, sizeof(struct ipq) == 200)
835 * We will commit 4MB at one time. Should we cross that limit
836 * we will prune down to 3MB, making room for approx 8 big 64K
839 net->ipv4.frags.high_thresh = 4 * 1024 * 1024;
840 net->ipv4.frags.low_thresh = 3 * 1024 * 1024;
842 * Important NOTE! Fragment queue must be destroyed before MSL expires.
843 * RFC791 is wrong proposing to prolongate timer each fragment arrival
846 net->ipv4.frags.timeout = IP_FRAG_TIME;
848 inet_frags_init_net(&net->ipv4.frags);
850 return ip4_frags_ns_ctl_register(net);
853 static void __net_exit ipv4_frags_exit_net(struct net *net)
855 ip4_frags_ns_ctl_unregister(net);
856 inet_frags_exit_net(&net->ipv4.frags, &ip4_frags);
859 static struct pernet_operations ip4_frags_ops = {
860 .init = ipv4_frags_init_net,
861 .exit = ipv4_frags_exit_net,
864 void __init ipfrag_init(void)
866 ip4_frags_ctl_register();
867 register_pernet_subsys(&ip4_frags_ops);
868 ip4_frags.hashfn = ip4_hashfn;
869 ip4_frags.constructor = ip4_frag_init;
870 ip4_frags.destructor = ip4_frag_free;
871 ip4_frags.skb_free = NULL;
872 ip4_frags.qsize = sizeof(struct ipq);
873 ip4_frags.match = ip4_frag_match;
874 ip4_frags.frag_expire = ip_expire;
875 ip4_frags.secret_interval = 10 * 60 * HZ;
876 inet_frags_init(&ip4_frags);